Wyecliff
All editions

Google Loses It's Best Engineer | Wyecliff Weekly #35

August 7, 20265 min read
The AI stories that actually matter this week | Edition #35 | August 7, 2026
This week the map of who runs AI got redrawn, and it had almost nothing to do with a new model. Google's AI brain trust broke apart, with its Nobel-winning chief stepping back and a 27-year veteran walking out the door. Meta launched its first tool that writes software on its own. And Apple and OpenAI took their fight over stolen secrets to a judge. Let's get to it.

Now Available: The Wyecliff Weekly Podcast

Prefer to listen? Every edition of the Wyecliff Weekly is now generated as a podcast. Same stories, same analysis, same lessons. Just press here.
Listen to this edition

Top Stories

+ Google Shook Up The Team Behind Its AI, And Its Most Famous Engineer Walked Out

On August 6, Google's parent company Alphabet announced a major overhaul of the group that builds its AI. Demis Hassabis, the Nobel Prize winner who ran Google DeepMind, is stepping back from running the day-to-day operation to become Alphabet's chief scientist and DeepMind's chairman. Koray Kavukcuoglu takes over the senior operating role. The bigger shock was who left: Jeff Dean, one of the most influential engineers in Google's history, walked away after 27 years, along with longtime partner Sanjay Ghemawat, to start a new company called Discovery Loop with two other senior Google researchers. The reshuffle lands at a rough moment. The top version of Google's newest Gemini model was supposed to arrive in June and still has not shipped, and rivals Anthropic and OpenAI each poached a star Google researcher this summer. Alphabet's stock fell about 4 percent on the news.
What it all means: The people are the product in AI, and this week Google lost some of its most important ones while shuffling the rest. A company that has been in the AI race since the beginning is now fighting to hold its team together. When the talent moves, the lead can move with it.

+ Meta Launched Its First AI That Writes Software On Its Own

On August 5, Meta released Muse Code, its first AI tool built to do the work of a software engineer with little human help. You describe what you want built, and it plans the changes, writes the code, checks its own work, and can even hand off pieces of the job to smaller helper programs it runs itself. It launched in a test version for developers, paired with an upgraded version of Meta's own coding AI, and it puts Meta head to head with Anthropic and OpenAI, who have led this corner of the market. Meta is also using price to get attention. Alongside the normal pay-as-you-go option, it offered a much cheaper tier for anyone willing to let Meta use their code to train future models. The effort is run by Alexandr Wang, the executive Meta paid handsomely to lead its new superintelligence group.
What it all means: Tools that build software by themselves are becoming the product every big AI company wants to sell, because that is where a lot of the money is. Meta showing up with a cheap option puts pressure on what everyone else can charge. The cut-rate tier, paid for with your own code, is the kind of trade worth reading closely before you take it.

+ Apple Asked A Judge To Stop OpenAI From Using Its Secrets

The fight between Apple and OpenAI got sharper this week. Back on July 10, Apple sued OpenAI, claiming former Apple employees carried confidential information about Apple's hardware plans to OpenAI. This week Apple went further, asking a federal judge to immediately bar OpenAI and two named former employees from using or sharing that information, and to block OpenAI from building devices with it while the case plays out. Apple says as many as 11 more former staffers may have been involved. OpenAI pushed back hard and in public. It published a blog post titled "Apple is getting this wrong," called the lawsuit "careless, aggressive and oddly personal," and argued that Apple's own loose security, including a manager who logged into a departed engineer's personal cloud account, undercuts the claim that these secrets were ever carefully guarded.
What it all means: Two of the most powerful companies in tech are fighting over the people and ideas moving between them, and they are doing it in the open. OpenAI is widely believed to be building a hardware device, which is what makes Apple nervous. How the judge rules could set the ground rules for how freely talent and know-how move across the whole industry.

More Stories

Anthropic Hired A Former State Supreme Court Justice To Run Global Policy.

On August 4, Anthropic named Mariano-Florentino Cuéllar, a former California Supreme Court justice, as its Chief Global Affairs Officer. The hire signals how much of the AI fight is now happening in courtrooms and capitals, not just in the lab.

A Chip Startup Raised $312 Million To Build Faster AI Hardware.

London-based OLIX Computing raised $312 million at a $3.3 billion value for chips that use light instead of only electricity to run AI faster and with less power. The raise is a bet that the scramble for AI computing power is nowhere near over.

More Than Half Of This Year's Layoffs Now Cite AI.

As of August 5, one running tracker counted 322 layoff events in 2026 affecting more than 205,000 workers, with 54 percent of those events naming AI or automation as a factor. The augmented-reality company Magic Leap cut about 200 jobs on August 3.

Claude Added A Safety Layer For Big Companies.

On August 5, Anthropic gave business customers a new control that inspects what employees type into Claude before it reaches the AI, letting compliance teams block sensitive company information from leaving. It is a sign that the next phase of AI at work is about guardrails, not just features.

Why It Matters

This was a week about who runs AI and who owns it. The teams behind the tools reshuffled, new players muscled into each other's territory, and the lawyers stepped in to sort out who owns the ideas. Google lost some of the people who built its lead, Meta charged into a market Anthropic and OpenAI had to themselves, and Apple and OpenAI took their fight over talent and secrets to a judge. The scoreboard moved more than the technology did.

For You

The names behind the assistants on your phone are changing hands. The person many considered the smartest engineer at Google just left to start something new, and the future of Google's assistant looks less certain than it did a month ago. More companies piling into AI, including Meta with a cheaper option, tends to mean lower prices and more choice for you over time. And a week spent watching these companies poach each other's people and sue over secrets is a reminder that the tools in your pocket are the prize in a very high-stakes fight.

For Your Work

The ground under the tools you rely on is shifting, and not because of a new feature. Leadership at one of the biggest AI labs just turned over, and a major new competitor entered the software-building market at a cut-rate price. For a buyer that points one direction: more options and more leverage, and good reason not to wire your whole operation to a single provider. The Apple and OpenAI fight is also a quiet lesson for anyone whose people move between companies, because how you handle confidential information on the way in and out is now the kind of thing that ends up in front of a judge.

One Thing To Try This Week

How to Show AI a Photo or Screenshot and Get Answers

Last week we turned on voice so you could talk to your AI out loud. This week we go the other direction: instead of describing something in words, you just show it. Every major assistant can now look at a photo or a screenshot and tell you what is in it, pull out the details, and answer questions about it. It is one of the most useful and least-used things these tools can do.

Claude (via Cowork or claude.ai)

  1. Open a chat at claude.ai or in the Claude app. This works on every plan, including free.
  2. Click the plus or paperclip icon in the message box to attach an image, or simply drag and drop the picture into the chat. Claude accepts common photo types like JPEG and PNG.
  3. Type your question about the image and send. You can attach more than one picture at a time and ask Claude to compare them.

ChatGPT

  1. Open ChatGPT on your phone or at chatgpt.com.
  2. On the phone, tap the plus icon and choose "Add photos & files," or take a new photo. On a computer, drag the image into the message box or paste one you have copied. It accepts PNG, JPEG, and still GIF images.
  3. Ask your question about the image and send. You can add more images later in the same conversation to dig deeper.

Microsoft Copilot

  1. Open the Copilot app or go to copilot.microsoft.com.
  2. Press the plus, or "Open," button and choose "Add images or files," then pick the picture from your device.
  3. Type your question about the image and send. Copilot will read the picture and answer, for example naming what is in a photo or pulling text out of a screenshot.
Bonus: This is a fix for dozens of small daily headaches. Snap a photo of an error message on your screen, a confusing bill, a chart in a report, a menu in another language, or a handwritten whiteboard, and ask the AI to explain it or type it up for you.
Try This Prompt
Here is a photo. Tell me what it shows, pull out the important details, and explain anything that looks unusual or that I should pay attention to. Then ask me if I want it turned into notes I can keep: [attach your image]

Spread The Word

Love the Wyecliff Weekly? Refer a friend and earn rewards on our referral program.

See the referral tiers
Wyecliff Weekly

Never miss an edition.

Wyecliff Weekly lands every Friday at 6 AM Mountain. One short email with the moves that matter, a prompt to run on Monday, and one playbook from the field.

One email a week. No spam. Unsubscribe anytime.

Common Questions

When does Wyecliff Weekly arrive?
Every Friday morning at 6 AM Mountain. One email a week with the biggest AI stories, what they mean for your work, and one thing to try before Monday.
Is Wyecliff Weekly free?
Yes. Subscribe with your work email and read every edition at no cost. We write for operators who want plain takes, not a sales pitch.
What is inside each edition?
Top AI news with context, a short take on why it matters for your business, and one practical prompt or workflow you can run this week.
How do I unsubscribe?
Every email includes an unsubscribe link at the bottom. One click and you are off the list. No hoops.
/newsletter/wyecliff-weekly-edition-35